Thomas A. Cwik

Status and directions for the PYRAMID parallel unstructured AMR library

This is a status report on our progress with the development of PYRAMID, a Fortran 90/95-based library for parallel unstructured adaptive mesh refinement. The library has been designed to simplify the use of adaptive methods in computational science applications by introducing many advanced software engineering features. In this paper, design and performance issues are described concluding with a discussion of our future development plans.

Computation of Radar Scattering From Heterogeneous Rough Soil Using the Finite-Element Method

A 2-D vector-element-based finite-element method (FEM) is used to calculate the radar backscatter from 1-D bare rough soil surfaces which can have an underlying heterogeneous substrate. Monte Carlo simulation results are presented for scattering at L-band (λ = 0.24 m). For homogeneous soils with rough surfaces, the results of FEM are compared with the predictions of the small perturbation method. In the case of heterogeneous substrates, soil moisture (and, hence, soil permittivity) is assumed to vary as a function of depth. In this case, the results of FEM are compared with those of the transfer matrix method for flat soil surfaces. In both cases, a good agreement is found. For homogeneous rough soils, it is found that polarimetric radar backscatter and copolarized phase difference have a nonlinear relationship with soil moisture. Finally, it is found that the nature of the soil moisture variation in the top few centimeters of the soil has a strong influence on the backscatter and, hence, on the inferred soil moisture content.

Early experiences with the myricom 2000 switch on an SMP Beowulf-class cluster for unstructured adaptive meshing

We explore the current capabilities of the recently released Myricom 2000 switch, using MPICH-GM for communication, on a 2-way SMP Pentium III Beowulf-Class cluster. Performance measurements indicate that data transfer rates of approximately 225 Mbytes/s with 9.3 microseconds latency for ping-pong tests can be achieved for messages as large as 32 Mbytes. When shared-memory communication is used approximately 130 MBytes/s with 1.5 microseconds latency for long messages (250 MBytes/s peak) is possible. The performance varies depending on how processors communicate; either within an SMP node or across nodes. Performance for parallel unstructured adaptive refinement of 3D tetrahedral meshes shows noticeable improvement when compared to 100BaseT Ethernet. Furthermore, when compared to traditional systems such as the SGI Origin 2000, the combination of this fast network with high performance SMP processors demonstrate that Beowulf-Clusters compare favorably with such systems--even for communication intensive applications.

Genetically Engineered Nanostructure Devices

Material variations on an atomic scale enable the quantum mechanical functionality of devices such as resonant tunneling diodes (RTDs), quantum well infrared photodetectors (QWIPs), quantum well lasers, and heterostructure field effect transistors (HFETs). The design and optimization of such heterostructure devices requires a detailed understanding of quantum mechanical electron transport. The Nanoelectronic Modeling Tool (NEMO) is a general-purpose quantum device design and analysis tool that addresses this problem. NEMO was combined with a parallelized genetic algorithm package (PGAPACK) to optimize structural and material parameters. The electron transport simulations presented here are based on a full band simulation, including effects of non-parabolic bands in the longitudinal and transverse directions relative to the electron transport and Hartree charge self-consistency. The first, result of the genetic algorithm driven quantum transport calculation with convergence of a random structure population to experimental data is presented. Introduction The NASA/JPL goal to reduce payload in future space missions while increasing mission capability demands miniaturization of measurement, analytical and communication systems. Currently, typical system requirements include the detection of particular spectral lines, associated data processing, and communication of the acquired data to other subsystems. While silicon device technology dominates the commercial microprocessor and memory market, semiconductor heterostructure devices maintain their niche for light detection, light emission, and high-speed data transmission. The production of these heterostructure devices is enabled by the advancement of material growth techniques, which opened up a vast design space. The full experimental exploration of this design space is unfeasible and a reliable design tool is needed. Military applications have similar system requirements to those listed above. Such requirements prompted a device modeling project at the Central Research Laboratory of Texas Instruments (which transferred to Raytheon Systems in 1997). NEMO was developed as a general-purpose quantum mechanics-based l-D device design and analysis tool from 1993-97. The tool is available to US researchers by request on the NEMO web site. NEMO is based on the nonequilibrium Green function approach, which allows a fundamentally sound inclusion of the required physics: bandstructure, scattering, and charge self-consistency. The theoretical approach is documented in references [2, 31 while some of the major simulation results are documented in references [4-61. This paper highlights the recent work on genetic algorithm based device parameter optimization. Quantum Device Parameter Optimization using Genetic Algorithms Heterostructure device designs involve the choice of material compositions, layer thicknesses, and oping profiles. Material parameters such as band offsets, effective masses, dielectric constants etc. influence the device simulation results in addition to the structural design parameters. The I Figure 1: Architecture of a genetic algorithm-based NEMO simulation. full exploration of the design space using purely experimental techniques is unfeasible due to time and financial constraints. For example, it takes a well-equipped research laboratory approximately five working days7 for the growth, processing and testing of a particular resonant tunneling diode design. NEMO can provide q~ant i ta t ive~-~ current voltage characteristics (I-Vs) within minutes to hours of CPU time for a single set of device and material parameters. With this quantitative simulation capability the design parameter space can be explored expediently once an automated system for the design parameter variation is implemented. This paper presents the combination of NEMO with a parallelized genetic algorithm package (PGAPACK) as indicated in Figure 1. The architecture lends itself to the optimization of any parameters that enter a NEMO simulation. To evaluate how good a particular parameter set is, a fitness function must be developed as discussed in the next section. Simulation Target and Fitness Function In this work the RTD is used as a vehicle to study the effects of structural and doping variations on the electron transport. I-Vs of two devices that are part of a well-behaved test matrix of experimental data published in reference [ 5 ] are used as a design target. The raw I-V data (see the example in Figure 2) contains a contact series resistance and oscillations in the negative differential resistance (NDR). The oscillation in the NDR is attributed to external circuit effects and cannot be simulated within NEMO. The step-like feature in the NDR is cut out of the raw data to generate a clean set of experimental data. The contact series resistance can be estimated from the peak current of a series of nominally identical devices with different cross sections. The voltage drop over the contact resistance can be subtracted out of the extrinsic voltage scale to yield the intrinsic voltage scale (see inset of Figure 2a) The fitness of the simulated data is measured against such target I-V. There are four particular features that are explicitly evaluated for each simulated I-V: peak and valley current and voltage, and the slope close to the peak and the valley (see Figure 2b). Differences between the target and the simulation in these four features and the absolute and relative error for all simulated data points enter into the fitness function with a weighted average. The target fitness evaluated against

Genetically engineered nanoelectronics

The quantum mechanical functionality of nanoelectronic devices such as resonant tunneling diodes (RTDs), quantum well infrared photodetectors (QWIPs), quantum well lasers, and heterostructure field effect transistors (HFETs) is enabled by material variations on an atomic scale. The design and optimization of such devices requires a fundamental understanding of electron transport in such dimensions. The nanoelectronic modeling tool (NEMO) is a general-purpose quantum device design and analysis tool based on a fundamental non-equilibrium electron transport theory. NEMO was combined with a parallelized genetic algorithm package (PGAPACK) to evolve structural and material parameters to match a desired set of experimental data. A numerical experiment that evolves structural variations such as layer widths and doping concentrations is performed to analyze an experimental current voltage characteristic. The genetic algorithm is found to drive the NEMO simulation parameters close to the experimentally prescribed layer thicknesses and doping profiles. With such a quantitative agreement between theory and experiment design synthesis can be performed.

A mesh reconfiguration scheme for speeding up Monte Carlo simulations of electromagnetic scattering by random rough surfaces

Abstract Traditional methods of Monte Carlo simulations of random rough surface scattering that use the finite element method involve the generation of multiple meshes for the purpose of taking ensemble averages. We propose a mesh reconfiguration scheme that instead uses a single master mesh. The main idea is to locally modify only the air–surface interface region in the mesh for each instance of a random rough surface. This method achieves a four fold improvement in computation time without any loss of accuracy.

A Robust and Scalable Library for Parallel Adaptive Mesh Refinement on Unstructured Meshes

The design and implementation of a software library for parallel adaptive mesh refinement in unstructured computations on multiprocessor systems are described. This software tool can be used in parallel finite element or parallel finite volume applications on triangular and tetrahedral meshes. It contains a suite of well-designed and efficiently implemented modules that perform operations in a typical P-AMR process. This includes mesh quality control during successive parallel adaptive mesh refinement, typically guided by a local-error estimator, and parallel load-balancing. Our P-AMR tool is implemented in Fortran 90 with a Message-Passing Interface (MPI) library, supporting code efficiency, modularity and portability. The AMR schemes, Fortran 90 data structures, and our parallel implementation strategies are discussed in the paper. Test results of our software, as applied to a few selected engineering finite element applications, will be demonstrated. Performance results of our code on Cray T3E, HP/Convex Exemplar systems, and on a PC cluster (a Beowulf-class system) will also be reported.

Using polarimetric sar data to infer soil moisture from surfaces with varying subsurface moisture profiles

A time-series approach is used to estimate the moisture content based on polarimetric SAR data. It is found that under the assumption of constant soil moisture, empirically observed relationships between radar backscatter and moisture are only half as sensitive to moisture as compared to actual radar data. A numerical finite element method is used to calculate the radar backscatter for rough soils with arbitrarily varying soil moisture as a function of depth. Several instance of drying and wetting moisture profiles are considered and the radar backscatter is calculated in each case. Radar backscatter is found to crucially depend on the soil moisture variation in the top half wavelength of soil.

Parallel Unstructured AMR and Gigabit Networking for Beowulf-Class Clusters

The impact of Gigabit networking with Myrinet 2000 hardware and MPICH-GM software on a 2-way SMP Beowulf-Class cluster for parallel unstructured adaptive mesh refinement using the PYRAMID library is described. Network performance measurements show that approximately 225 Mbytes/s (1.8 Gbits/s) with 9.3µ seconds latency can be achieved for large messages in ping-pong tests. The performance varies depending on how processors communicate; either within an SMP node or across nodes. When applied to parallel unstructured AMR of 3D tetrahedral meshes noticeable improvement is observed when compared to 100BaseT (100 Mbits/s) Ethernet. Furthermore, when compared to traditional systems, such as the SGI Origin 2000, one realizes that Beowulf-Clusters with fast networks and high performance SMP processors compare favorably with such systems-even for communication intensive applications.

Genetically engineered microelectronic infrared filters

A genetic algorithm is used for design of infrared filters and in the understanding of the material structure of a resonant tunneling diode. These two components are examples of microdevices and nanodevices that can be numerically simulated using fundamental mathematical and physical models. Because the number of parameters that can be used in the design of one of these devices is large, and because experimental exploration of the design space is unfeasible, reliable software models integrated with global optimization methods are examined. The genetic algorithm and engineering design codes have been implemented on massively parallel computers to exploit their high performance. Design results are presented for the infrared filter showing new and optimized device design.